Programming Project #1: Hybrid Images¶
CS445: Computational Photography¶
Part I: Hybrid Images¶
In [1]:
import cv2
import numpy as np
from matplotlib.colors import LogNorm
import matplotlib.pyplot as plt
from scipy import signal
import sys
import utils
In [2]:
# switch from notebook to inline if using colab or otherwise cannot use interactive display)
%matplotlib notebook
import matplotlib.pyplot as plt
In [3]:
# im1_file = datadir + 'nutmeg.jpg'
# im2_file = datadir + 'DerekPicture.jpg'
# im1_file = './img/LecExample/Nutmeg.jpg'
# im2_file = './img/LecExample/DerekPicture.jpg'
im1_file = './img/set1/lama.jpg'
im2_file = './img/set1/owl.jpg'
# im1_file = './img/extra/1_wolf.jpg'
# im2_file = './img/extra/1_geralt.jpeg'
# im1_file = './img/extra/2_owl.jpeg'
# im2_file = './img/extra/2_cat.jpg'
im1 = np.float32(cv2.imread(im1_file, cv2.IMREAD_GRAYSCALE) / 255.0)
im2 = np.float32(cv2.imread(im2_file, cv2.IMREAD_GRAYSCALE) / 255.0)
im21 = cv2.imread(im1_file, cv2.IMREAD_GRAYSCALE)
im22 = cv2.imread(im2_file, cv2.IMREAD_GRAYSCALE)
# print(im21)
In [5]:
pts_im1 = utils.prompt_eye_selection(im1)
# pts_im1 = np.array([[607, 290], [748, 370]]) # uncomment if entering [x, y] pts manually
# plt.plot(pts_im1[:,0], pts_im1[:,1], 'r-+')
In [6]:
pts_im2 = utils.prompt_eye_selection(im2)
# pts_im2 = np.array([[299,343], [439,331]]) # uncomment if entering [x, y] pts manually
# plt.plot(pts_im2[:,0], pts_im2[:,1], 'r-+')
In [7]:
im1, im2 = utils.align_images(im1_file, im2_file,pts_im1,pts_im2,save_images=False)
In [8]:
# convert to grayscale
im1 = cv2.cvtColor(im1, cv2.COLOR_BGR2GRAY) / 255.0
im2 = cv2.cvtColor(im2, cv2.COLOR_BGR2GRAY) / 255.0
In [9]:
#Images sanity check
fig, axes = plt.subplots(1, 2)
axes[0].imshow(im1,cmap='gray')
axes[0].set_title('Image 1'), axes[0].set_xticks([]), axes[0].set_yticks([])
axes[1].imshow(im2,cmap='gray')
axes[1].set_title('Image 2'), axes[1].set_xticks([]), axes[1].set_yticks([]);
In [10]:
#Get FFT magnitude of im1 and im2
fftmag1 = np.abs(np.fft.fftshift(np.fft.fft2(im1)))
fftmag2 = np.abs(np.fft.fftshift(np.fft.fft2(im2)))
fig, axes = plt.subplots(1, 2)
#fft display here
axes[0].imshow(fftmag1,norm=LogNorm(fftmag1.min(),fftmag1.max()),cmap='jet')
axes[0].set_title('FFT of im1'), axes[0].set_xticks([]), axes[0].set_yticks([])
axes[1].imshow(fftmag2,norm=LogNorm(fftmag2.min(),fftmag2.max()),cmap='jet')
axes[1].set_title('FFT of im2'), axes[1].set_xticks([]), axes[1].set_yticks([]);
# plt.savefig('./img/set1/fft1')
# plt.savefig('./img/set1/fft2')
In [11]:
#we get a gaussian filter here, for low pass
sigmaVal = 40
def getGaussian(sigma):
#sigma decides how large is the kernel
ksize = np.int(np.ceil(sigma)*6+1)
gaussianFilter = cv2.getGaussianKernel(ksize, sigma) # 1D kernel
gaussianFilter = gaussianFilter*np.transpose(gaussianFilter) # 2D kernel by outer product
return gaussianFilter
# fig, axes = plt.subplots(1, 2)
# axes[0].imshow(getGaussian(sigmaVal),cmap='gray')
# axes[0].set_title('gaussianFilter'), axes[0].set_xticks([]), axes[0].set_yticks([])
#for high pass, use impulse filter minus gaussian
# impluse = signal.unit_impulse(gaussianFilter.shape, 'mid')
lowPass = getGaussian(sigmaVal/10)
# axes[1].imshow(lowPass,cmap='gray')
# axes[1].set_title('low pass'), axes[1].set_xticks([]), axes[1].set_yticks([]);
In [12]:
def hybridImage(im1, im2, sigma_low, sigma_high):
Filter_cutlow = getGaussian(sigma_low)
Filter_cuthigh = getGaussian(sigma_high)
im_fil1 = cv2.filter2D(im1, -1, Filter_cutlow)
plt.figure()
plt.imshow(im_fil1, cmap='gray')
# plt.savefig('./img/set1/im_fil1')
im_fil2 = im2 - cv2.filter2D(im2, -1, Filter_cuthigh)
plt.figure()
plt.imshow(im_fil2,cmap='gray')
# plt.savefig('./img/set1/im_fil2')
fftmag1 = np.abs(np.fft.fftshift(np.fft.fft2(im_fil1)))
plt.figure()
plt.imshow(fftmag1, norm=LogNorm(1, 30))
# plt.savefig('./img/set1/fftmag1')
fftmag2 = np.abs(np.fft.fftshift(np.fft.fft2(im_fil2)))
plt.figure()
plt.imshow(fftmag2, norm=LogNorm(1, 30))
# plt.savefig('./img/set1/fftmag2')
res = (im_fil1 + im_fil2)/2
plt.figure()
plt.imshow(res,cmap='gray')
# plt.savefig('./img/set1/res')
# plt.savefig('./img/extra/extraRes2')
fftres = np.abs(np.fft.fftshift(np.fft.fft2(res)))
plt.figure()
plt.imshow(fftres,norm=LogNorm(1, 30))
# plt.savefig('./img/set1/fftres')
return res
In [15]:
sigma_low = 3 # choose parameters that work for your images
sigma_high = 2
res = hybridImage(im2, im1, sigma_low, sigma_high)
In [16]:
# Optional: Select top left corner and bottom right corner to crop image
# the function returns dictionary of
# {
# 'cropped_image': np.ndarray of shape H x W
# 'crop_bound': np.ndarray of shape 2x2
# }
cropped_object = utils.interactive_crop(res)
Part II: Image Enhancement¶
Two out of three types of image enhancement are required. Choose a good image to showcase each type and implement a method. This code doesn't rely on the hybrid image part.¶
Contrast enhancement¶
In [17]:
im_origin = cv2.imread('./img/part2/img2_1.jpg', 1)
#histogram equalization
#credit to
#https://opencv-python-tutroals.readthedocs.io/en/latest/py_tutorials/py_imgproc/py_histograms/py_histogram_equalization/py_histogram_equalization.html
#stack overflow
#https://stackoverflow.com/questions/31998428/opencv-python-equalizehist-colored-image
# RGB to HSV
img_hsv = cv2.cvtColor(im_origin, cv2.COLOR_RGB2HSV)
img_hsv[:, :, 2] = cv2.equalizeHist(img_hsv[:, :, 2])#Hue, Saturation, and Value, so 2 is v
#HSV to RGB
im_after = cv2.cvtColor(img_hsv, cv2.COLOR_HSV2RGB)
im_after = im_after[:, :, [2,1,0]]# maps BGR to RGB
im_origin = im_origin[:, :, [2,1,0]]
plt.figure()
plt.imshow(im_origin, norm=LogNorm(1, 30))
# plt.savefig('./img/part2/im_origin')
plt.figure()
plt.imshow(im_after, norm=LogNorm(1, 30))
# plt.savefig('./img/part2/im_after')
# axes[0].imshow(im_origin,norm=LogNorm(im_origin.min(),im_origin.max()),cmap='jet')
# axes[0].set_title('im_origin'), axes[0].set_xticks([]), axes[0].set_yticks([])
# axes[1].imshow(im_after,norm=LogNorm(im_after.min(),im_after.max()),cmap='jet')
# axes[1].set_title('im_origin'), axes[0].set_xticks([]), axes[0].set_yticks([])
# plt.imshow(res)
# plt.imshow(im_after)
Out[17]:
Color enhancement¶
In [18]:
#Color value refers to the relative lightness or darkness of a color.
# Note that you want the values to map between the range defined by the imported library
#(in OpenCv 0-255), so you shouldn't just add or multiply with some constant
def change_s(img, val):
#img_hsv = cv2.cvtColor(im_origin, cv2.COLOR_RGB2HSV)
hsv = cv2.cvtColor(img, cv2.COLOR_BGR2HSV)
h, s, v = cv2.split(hsv)
#do sth to s
#cv2.inrange
s = cv2.add(s, val)
s[s< 0] = 0
s[s>255] = 255
final_hsv = cv2.merge((h, s, v))
img = cv2.cvtColor(final_hsv, cv2.COLOR_HSV2BGR)
return img
img = cv2.imread('./img/part2/img2_1.jpg')
original = img[:, :, [2,1,0]]
plt.figure()
plt.imshow(original, norm=LogNorm(1, 30))
# plt.savefig('./img/part2/2_2im_origin')
img = change_s(img, 70) #increases
img = img[:, :, [2,1,0]]
plt.figure()
plt.imshow(img, norm=LogNorm(1, 30))
# plt.savefig('./img/part2/2_2color_enhance')
Out[18]:
Color shift¶
In [19]:
# Take an image of your choice and create two color-modified
# versions that are (a) more red; (b) less yellow.
#In OpenCv use cv2.cvtColor(image, cv2.COLOR_BGR2Lab)
# L – Lightness ( Intensity ).
# a – color component ranging from Green to Red.
# b – color component ranging from Blue to Yellow.
def colorShift(img,valRed, valYellow):
lab = cv2.cvtColor(img, cv2.COLOR_BGR2Lab)
# plt.imshow(lab, norm=LogNorm(1, 30))
l,a,b=cv2.split(lab)
#increase a, more res?
a = cv2.add(a, valRed)
#decrease b, less yellow?
b = cv2.add(a,valYellow)
# b[b< -127] = 127
# b[b>128] = 128
final_lab = cv2.merge((l, a, b))
img = cv2.cvtColor(final_lab, cv2.COLOR_LAB2BGR)
img = img[:, :, [2,1,0]]
return img
img = cv2.imread('./img/part2/part2_2.jpeg')
img = img[:, :, [2,1,0]]
plt.figure()
plt.title('original fig')
plt.imshow(img, norm=LogNorm(1, 30))
img_moreRed = colorShift(img,20,0)
plt.figure()
plt.title('More Red fig')
plt.imshow(img_moreRed, norm=LogNorm(1, 30))
# plt.savefig('./img/part2/moreRed')
img_lessYellow = colorShift(img,0,-10)
plt.figure()
plt.title('Less Yellow fig')
plt.imshow(img_lessYellow, norm=LogNorm(1, 30))
# plt.savefig('./img/part2/lessYellow')
Out[19]:
Part III: B & W¶
Enhance the hybrid img¶
In [ ]:
#Try using color to enhance the effect of hybrid images.
# Does it work better to use color for the high-frequency component,
# the low-frequency component, or both? (5 pts)
p3img = cv2.imread('./img/set1/res.png')
# plt.figure()
# plt.imshow(p3img)
newimg = change_s(p3img, 0)
newimg = newimg[:, :, [2,1,0]]
plt.figure()
plt.imshow(newimg, norm=LogNorm(1, 30))
#####not done
Gaussian and Laplacian pyramids¶
In [ ]:
# def deletes(img):
# tmpFig = np.copy(fig)
# row = img.shape[0]
# col = img.shape[1]
# channel = img.shape[2]
# # print(row, col, channel)
# for i in range channel:
# for j in range row:
# for k in range col:
# if(k % 2 == 0):
# return tmpFig
####abandon
In [20]:
# Illustrate the hybrid image process by implementing Gaussian and
# Laplacian pyramids and displaying them for your favorite result.
# This should look similar to Figure 7
# in the Oliva et al. paper. (15 pts)
img = cv2.imread('./img/part3/p3_pyr.png')
gFilter = Filter_cutlow = getGaussian(1)
initSize = 4
plt.figure(figsize=(initSize,initSize))
plt.imshow(img)
# store the high_frequence info
store = []
# print(img.shape)
for x in range(3):
old_img = img
img = cv2.filter2D(old_img, -1, gFilter)
high_freq = old_img - img
store.append(high_freq)
img = cv2.resize(img, # original image
None, # set fx and fy, not the final size
fx=0.5,
fy=0.5,
interpolation=cv2.INTER_LINEAR)
plt.figure(figsize=(initSize,initSize))
# print(img.shape)
plt.imshow(img)
for i in range(3):
print("check shape", i ,store[i].shape)
In [21]:
# A Laplacian pyramid is very similar to a Gaussian pyramid but saves the
# difference image of the blurred versions between each levels.
img = cv2.imread('./img/part3/p3_pyr.png')
bigSize = initSize
plt.figure(figsize=(bigSize,bigSize))
plt.imshow(img)
for i in range(len(store)):
plt.figure(figsize=(initSize,initSize))
# plt.imshow(store[i])
plt.imshow(store[i], norm=LogNorm(1, 20))
# print(len(store))
# for x in range(3):
# old_img = img
# img = cv2.filter2D(old_img, -1, gFilter)
# print("before",big_img.shape)
# big_img = cv2.resize(big_img, # original image
# None, # set fx and fy, not the final size
# fx=2.0,
# fy=2.0,
# )
# plt.figure(figsize=(bigSize,bigSize))
# plt.imshow(big_img)
# print("AFTER",big_img.shape)
In [ ]: